Merge pull request #1277 from 0-complexity/migrate-vfw

fix for migrating halted vfw

Author: alichaddad

apps/cbportal/base/cloudbroker__cloudspace/methodclass/cloudbroker_cloudspace.py

@@ -234,7 +234,7 @@ def _destroyVFW(self, gid, cloudspaceId, deletemodel=True):
         fws = self.cb.netmgr.fw_list(gid=int(gid), domain=str(cloudspaceId))
         if fws:
             try:
-                self.cb.netmgr.fw_delete(fwid=fws[0]['guid'], gid=gid, deletemodel=deletemodel, timeout=20)
+                self.cb.netmgr.fw_delete(fwid=fws[0]['guid'], deletemodel=deletemodel, timeout=20)
             except exceptions.ServiceUnavailable:
                 return False
             return True

apps/cbportal/base/cloudbroker__computenode/methodclass/cloudbroker_computenode.py

@@ -3,8 +3,6 @@
 import functools
 from JumpScale.portal.portal import exceptions
 from cloudbrokerlib.baseactor import BaseActor, wrap_remote
-import random
-
 
 class cloudbroker_computenode(BaseActor):
     """
@@ -191,9 +189,11 @@ def _move_virtual_machines(self, stack, title, ctx):
         vfws = self._vcl.search({'gid': stack['gid'],
                                  'nid': int(stack['referenceId'])})[1:]
         for vfw in vfws:
-            randomnode = random.choice(othernodes)
+            nid = int(self.cb.getBestProvider(stack['gid'], memory=128)['referenceId'])
             ctx.events.sendMessage(title, 'Moving Virtual Firewal %s' % vfw['id'])
-            self.cb.netmgr.fw_move(vfw['guid'], randomnode['id'])
+            if not self.cb.netmgr.fw_move(vfw['guid'], nid):
+                self.cb.netmgr.fw_delete(fwid=vfw['guid'], deletemodel=False, timeout=20)
+                self.cb.netmgr.fw_start(vfw['guid'], targetNid=nid)
 
     @auth(['level2', 'level3'], True)
     @wrap_remote

apps/cloudbroker/cloudbrokerlib/cloudbroker.py

@@ -547,7 +547,7 @@ def release_resources(self, cloudspace, releasenetwork=True):
             stack = next(iter(models.stack.search({'referenceId': str(fw.nid), 'gid': fw.gid})[1:]), None)
             if stack and stack['status'] != 'DECOMISSIONED':
                 # destroy vm and model
-                self.cb.netmgr.fw_delete(fwguid, cloudspace.gid)
+                self.cb.netmgr.fw_delete(fwguid)
             else:
                 # destroy model only
                 self.cb.netmgr.fw_destroy(fwguid)

apps/cloudbroker/cloudbrokerlib/netmgr.py

@@ -179,7 +179,7 @@ def fw_get_openvpn_config(self, fwid, **kwargs):
             raise exceptions.ServiceUnavailable("Failed to get OpenVPN Config")
         return job['result']
 
-    def fw_delete(self, fwid, gid, deletemodel=True, timeout=600, **kwargs):
+    def fw_delete(self, fwid, deletemodel=True, timeout=600, **kwargs):
         """
         param:fwid firewall id
         param:gid grid id

libs/agent-scripts/vfw/vfs_migrate_routeros.py

@@ -39,34 +39,38 @@ def action(networkid, sourceip, vlan, externalip):
     if source_con:
         templatepath = '/var/lib/libvirt/images/routeros/template/'
         destination = '/var/lib/libvirt/images/routeros/{0:04x}'.format(networkid)
-        domain = source_con.lookupByName(name)
-        if domain.state()[0] == libvirt.VIR_DOMAIN_RUNNING:
-            if not j.system.fs.exists(destination):
-                print 'Creating image snapshot %s -> %s' % (templatepath, destination)
-            localip = j.system.net.getReachableIpAddress(sourceip, 22)
-            targeturl = "tcp://{}".format(localip)
-            j.system.btrfs.snapshot(templatepath, destination)
-            xmldom = ElementTree.fromstring(domain.XMLDesc())
-            seclabel = xmldom.find('seclabel')
-            if seclabel is not None:
-                xmldom.remove(seclabel)
-            xml = ElementTree.tostring(xmldom)
-            flags = libvirt.VIR_MIGRATE_LIVE | libvirt.VIR_MIGRATE_PERSIST_DEST | libvirt.VIR_MIGRATE_UNDEFINE_SOURCE | libvirt.VIR_MIGRATE_NON_SHARED_DISK
-            try:
-                domain.migrate2(target_con, flags=flags, dxml=xml, uri=targeturl)
-            except:
+        try:
+            domain = source_con.lookupByName(name)
+        except libvirt.libvirtError:
+            domain = None
+        if domain:
+            if domain.state()[0] == libvirt.VIR_DOMAIN_RUNNING:
+                if not j.system.fs.exists(destination):
+                    print 'Creating image snapshot %s -> %s' % (templatepath, destination)
+                localip = j.system.net.getReachableIpAddress(sourceip, 22)
+                targeturl = "tcp://{}".format(localip)
+                j.system.btrfs.snapshot(templatepath, destination)
+                xmldom = ElementTree.fromstring(domain.XMLDesc())
+                seclabel = xmldom.find('seclabel')
+                if seclabel is not None:
+                    xmldom.remove(seclabel)
+                xml = ElementTree.tostring(xmldom)
+                flags = libvirt.VIR_MIGRATE_LIVE | libvirt.VIR_MIGRATE_PERSIST_DEST | libvirt.VIR_MIGRATE_UNDEFINE_SOURCE | libvirt.VIR_MIGRATE_NON_SHARED_DISK
                 try:
-                    target_domain = target_con.lookupByName(name)
-                    target_domain.undefine()
+                    domain.migrate2(target_con, flags=flags, dxml=xml, uri=targeturl)
                 except:
-                    pass  # vm wasn't created on target
-                raise
-            domain = target_con.lookupByName(name)
-            network.protect_external(domain, externalip)
-            network.protect_gwmgmt(domain, internalip)
-        else:
-            domain.undefine()
-            return False
+                    try:
+                        target_domain = target_con.lookupByName(name)
+                        target_domain.undefine()
+                    except:
+                        pass  # vm wasn't created on target
+                    raise
+                domain = target_con.lookupByName(name)
+                network.protect_external(domain, externalip)
+                network.protect_gwmgmt(domain, internalip)
+            else:
+                domain.undefine()
+                return False
         # remove disk from source
         con = j.remote.cuisine.connect(sourceip, 22)
         con.run('btrfs subvol delete {} || true'.format(destination))

libs/agent-scripts/vfw/vfs_routeros_restore.py

@@ -27,19 +27,20 @@ def action(networkid):
     print('Restoring {} to {}'.format(ovslocation, localfile))
     destination = '/var/lib/libvirt/images/routeros/'
     networkidHex = '%04x' % int(networkid)
-    if j.system.fs.exists(j.system.fs.joinPaths(destination, networkidHex)):
-        j.system.btrfs.subvolumeDelete(destination, networkidHex)
-    j.system.btrfs.subvolumeCreate(destination, networkidHex)
     try:
         j.system.platform.qemu_img.info(ovslocation)
     except RuntimeError as e:
         if 'No such file or directory' in e.message:
             return False
         j.errorconditionhandler.processPythonExceptionObject(e)
         return False
+    if j.system.fs.exists(j.system.fs.joinPaths(destination, networkidHex)):
+        j.system.btrfs.subvolumeDelete(destination, networkidHex)
+    j.system.btrfs.subvolumeCreate(destination, networkidHex)
     try:
         j.system.platform.qemu_img.convert(ovslocation, 'raw', localfile, 'qcow2')
     except Exception as e:
+        j.system.btrfs.subvolumeDelete(destination, networkidHex)
         j.errorconditionhandler.processPythonExceptionObject(e)
         return False
     return True

scripts/ops/releaseCloudSpaceResources.py

@@ -25,7 +25,7 @@ def _release_resources(cloudspaceId):
     # delete routeros
     fws = netmgr.fw_list(gridid, str(cloudspace.id))
     if fws:
-        netmgr.fw_delete(fws[0]['guid'], gridid)
+        netmgr.fw_delete(fws[0]['guid'])
     if cloudspace.networkId:
         libvirt_actor.releaseNetworkId(cloudspace.networkId)
     if cloudspace.publicipaddress: